🍓 Strawberry Tech Notes

Think about the last time you played a PC game, trained an AI model, generated an image, or asked an AI assistant to write some code.

There's a good chance NVIDIA technology was involved somewhere along the way.

Today, NVIDIA is one of the most influential companies in the technology industry. Its GPUs power AI training, data centres, scientific research, autonomous driving, robotics, and much more.

But NVIDIA didn't start as an AI company.

It started with a much narrower ambition. "Making computer graphics better".

So how did a company known for gaming graphics become one of the central players in the global AI revolution?

The answer involves a GPU, a decision made years before the AI boom, a software platform called CUDA, and a completely different way of thinking about computing.

Let's go back to the beginning.

1. Where It All Started

NVIDIA was founded on April 5, 1993, by three engineers.

• Jensen Huang
• Chris Malachowsky
• Curtis Priem

Their original vision was to develop advanced graphics technology for the PC gaming and multimedia markets.

At the time, computer graphics were becoming increasingly important. Games were becoming more visually complex, and traditional processors weren't designed specifically to handle the enormous number of calculations required for real-time graphics.

NVIDIA saw an opportunity.

Its founders believed there was a market for specialised hardware that could make graphics processing significantly faster.

Today, Jensen Huang remains NVIDIA's CEO and has led the company since its founding.

What's particularly interesting is that NVIDIA's journey wasn't a straight line to success. The company faced serious challenges during its early years, including difficult product decisions and intense competition.

It had to adapt, make difficult bets, and repeatedly find its place in a rapidly changing technology industry.

One of its most important milestones arrived in 1999.

2. The GPU: More Than Just Better Graphics

In 1999, NVIDIA introduced the GeForce 256, which it marketed as the world's first GPU (Graphics Processing Unit).

But *what exactly is a GPU, and why was it such an important invention?
*
To understand that, let's compare two types of processors.

CPU vs GPU

A CPU (Central Processing Unit) is designed to handle a wide variety of tasks.

It generally has a smaller number of powerful cores, each capable of handling complex instructions and different types of operations.

A GPU, on the other hand, is designed to handle large numbers of similar operations in parallel.

For example, imagine rendering a computer game.

A screen contains millions of pixels. Each pixel may require calculations involving colour, lighting, textures, shadows, and other visual effects.

Instead of processing every calculation one after another, a GPU can distribute many similar calculations across its processing units.

That's the power of parallel computing.

Here's a simplified comparison.

CPU

• Fewer, more versatile cores
• Designed for a wide range of tasks
• Strong at complex, sequential operations and coordinating work

GPU

• Many processing units designed for parallel workloads
• Excellent at performing many similar calculations simultaneously
• Particularly useful for graphics, scientific computing, and AI workloads

This doesn't mean CPUs only process tasks sequentially or GPUs can only perform parallel calculations. Modern processors support both approaches, but they're optimised for different workloads.

Why does this matter for AI?

Training an AI model involves enormous amounts of mathematical computation, particularly matrix and tensor operations.

These operations can often be divided into smaller calculations and executed in parallel.

That makes GPUs particularly useful for AI workloads.

A technology originally developed to render video game graphics turned out to be extremely valuable for something much bigger.

But NVIDIA still needed a way for developers to use GPUs for more than graphics.

That's where its next major decision comes in.

3. CUDA: The Decision That Changed Everything

In 2006, NVIDIA introduced CUDA (Compute Unified Device Architecture).

This was arguably one of the most important decisions in the history of company.

Before CUDA, GPUs were primarily associated with graphics processing.

CUDA opened up GPU parallel-processing capabilities to general-purpose computing.

In other words, developers could now use NVIDIA GPUs for workloads that had nothing to do with rendering video games.

These included,

• Scientific simulations
• Financial modelling
• Medical imaging
• Data processing
• Machine learning
• Artificial intelligence

But what exactly is CUDA?

CUDA is not simply a GPU or a programming language.

It's a parallel computing platform and programming model that allows developers to write software that takes advantage of NVIDIA GPUs.

It includes programming tools, libraries, APIs, and other development resources.

For example, a developer can write GPU-accelerated code using CUDA C++.

Instead of performing every calculation on the CPU, the application can send suitable workloads to the GPU.

A simplified example.

__global__ void add(int *a, int *b, int *c) {
int i = threadIdx.x;
c[i] = a[i] + b[i];
}

This is a small CUDA kernel. It performs an addition operation on data using GPU execution threads.

The example is deliberately simplified, but it illustrates the basic idea" developers can write code that runs on the GPU".

Why was CUDA such a big deal?

Imagine two companies selling powerful processors.

Both have impressive hardware.

But one company also has

• Mature development tools
• Extensive documentation
• Optimised libraries
• Developer communities
• Existing applications
• Years of software compatibility

Which platform would developers find easier to adopt?

This is where NVIDIA gained a major advantage.

CUDA created an ecosystem around its hardware.

Over time, developers built applications and tools specifically designed to work with NVIDIA GPUs.

For companies already using that ecosystem, switching to a different platform could mean more than purchasing a different chip. It could require adapting software, changing infrastructure, and optimising existing workloads.

This is often described as a software ecosystem advantage or a competitive moat.

And NVIDIA started building that advantage long before generative AI became mainstream.

4. The Moment AI Changed NVIDIA Future

NVIDIA GPUs were already being used in scientific computing and machine learning.

But an important moment arrived in 2012.

A deep-learning model called AlexNet achieved a breakthrough result in the ImageNet image-recognition competition.

The model was trained using NVIDIA GPUs.

This helped demonstrate how GPU acceleration could make large-scale deep-learning workloads practical.

Why was this important?

Deep learning involves neural networks with many layers and enormous numbers of mathematical operations.

GPUs can accelerate many of these operations, making it possible to train increasingly complex models.

The success of AlexNet helped accelerate interest in GPU-powered deep learning.

NVIDIA was no longer just building hardware for better graphics.

Its technology was becoming part of the infrastructure behind modern AI research.

Then came another major shift.

Generative AI

Large language models, image-generation systems, and other AI applications created an enormous demand for computing power.

Training and serving these models required powerful accelerators, high-bandwidth memory, fast networking, and large-scale data-centre infrastructure.

NVIDIA was already positioned to supply many of these components.

5. Why NVIDIA GPUs Are So Important for AI

Let's connect this to something many developers are familiar with Large Language Models.

When an LLM is trained, the system performs enormous numbers of mathematical operations involving tensors and matrices.

These operations are highly parallelisable.

NVIDIA GPUs are designed to accelerate this type of computation.

Modern NVIDIA data-centre GPUs also include specialised hardware, such as Tensor Cores, designed to accelerate matrix operations commonly used in AI.

But there's more to AI performance than simply having a powerful GPU.

Other important components include,

• GPU memory capacity
• Memory bandwidth
• Interconnect speed
• Networking between GPUs
• Power consumption
• Software optimisation
• Cooling and data-centre infrastructure

For example, training a large model may require thousands of GPUs working together.

Those GPUs need to communicate and exchange data efficiently.

A collection of powerful GPUs without effective communication and coordination wouldn't deliver the same performance as a properly designed computing system.

This is why NVIDIA's business has expanded far beyond selling individual graphics cards.

6. From Selling GPUs to Building AI Infrastructure

NVIDIA increasingly sells complete computing platforms rather than focusing only on individual chips.

Its data-centre solutions combine several technologies.

🍓 Blackwell

Blackwell is one of the major AI computing architectures of NVIDIA.

It was designed to support demanding AI workloads, including large-scale model training and inference.

The Blackwell platform includes GPUs, CPUs, networking, and systems designed to work together.

🍓 Vera Rubin

NVIDIA's next major platform is Vera Rubin.

It combines Rubin GPUs with NVIDIA's Vera CPUs and other components to build large-scale AI computing systems.

The platform is designed to support increasingly demanding AI workloads, including agentic AI and advanced reasoning models.

In May 2026, NVIDIA announced that Vera Rubin was ramping into full production, with partners preparing to deploy systems at scale.

This is an important distinction. An announced architecture, a platform entering production, and a system available to customers are different stages of a product's lifecycle.

NVIDIA has also outlined future roadmap generations, including Rubin Ultra and Feynman.

The broader strategy is clear.

NVIDIA wants to keep advancing computing performance while reducing the cost and energy required to generate AI output.

7. What Is an AI Factory?

You've probably heard the term data centre.

But NVIDIA increasingly describes its large-scale AI infrastructure as an AI factory.

Why?

A traditional data centre stores, processes, and delivers information.

An AI factory uses computing infrastructure to process data, train models, and generate intelligence.

Think about an application such as ChatGPT.

Every time someone sends a prompt, the system needs computing resources to process the request and generate a response.

At scale, millions of users can generate enormous numbers of requests.

The infrastructure behind those interactions needs to handle.

• Model inference
• Data processing
• GPU scheduling
• Networking
• Storage
• Power management
• Cooling

NVIDIA's vision is to provide the computing platform that powers these AI factories.

NVLink and NVL72

One example is NVIDIA NVL72 rack-scale system.

**Rather than treating each GPU as an isolated component, it connects large numbers of GPUs using high-speed interconnect technology.

NVLink enables high-bandwidth communication between GPUs.**

NVIDIA also develops networking technologies such as InfiniBand and Spectrum-X Ethernet to connect systems across larger AI clusters.

The result is an integrated computing infrastructure designed to operate as one large AI system.

This is a significant change in how NVIDIA positions itself.

It's not just selling chips.

It's selling the infrastructure needed to build and operate AI systems.

8. NVIDIA Beyond Generative AI

AI infrastructure is a major part of NVIDIA business, but its technology is also being applied in several other industries.

NVIDIA Omniverse

Omniverse is a platform for building and connecting 3D simulations and digital twins.

A digital twin is a virtual representation of a real-world system.

For example, a manufacturer could create a digital version of a factory to simulate production processes before changing the physical environment.

This can help engineers.

• Test factory layouts
• Simulate machinery
• Analyse production workflows
• Identify potential problems
• Experiment with changes before implementing them in the real world

Robotics and Physical AI

NVIDIA is also working on technologies for robotics and physical AI.

Physical AI refers to AI systems that interact with and operate in the physical world.

Examples include,

• Industrial robots
• Autonomous machines
• Humanoid robots
• Self-driving vehicles

Simulation environments can help train and test these systems before they operate in real-world conditions.

NVIDIA's Cosmos platform focuses on world models and physical AI development, while its Isaac platform provides technologies for robotics.

Automotive

NVIDIA also develops DRIVE platforms for autonomous driving and intelligent vehicle systems.

These technologies combine computing hardware, software, and AI capabilities to support advanced driver-assistance and autonomous-driving development.

The common theme across these industries is accelerated computing.

NVIDIA is trying to make its platform useful wherever large amounts of computation are required.

9. How Big Has NVIDIA Become?

Now let's look at the business transformation.

NVIDIA financial growth during the AI boom has been extraordinary.

For fiscal year 2026, which ended on January 25, 2026, NVIDIA reported,

-** Revenue:** $215.9 billion

• Net income: $120.1 billion
• Revenue growth: 65% compared with the previous fiscal year
• Data Centre revenue: $193.7 billion
• Gaming revenue: $16 billion

The numbers reveal just how much the business has changed.

Data Centre revenue accounted for approximately 90% of total revenue.

Gaming, the market that helped establish NVIDIA, represented a much smaller share.

The company that began by building graphics technology for games had become a major supplier of global AI infrastructure.

It's also worth distinguishing between revenue and profit.

Revenue is the money a company earns from selling its products and services.

Net income is what's left after expenses, taxes, and other relevant costs.

NVIDIA's fiscal 2026 net income exceeded $120 billion, showing just how profitable its business had become during this period.

These figures are from NVIDIA's reported financial results, not a prediction or market estimate.

10. Why Doesn't NVIDIA Manufacture Its Own Chips?

Here's something interesting about NVIDIA's business model.

NVIDIA designs its chips, but it doesn't operate the enormous semiconductor fabrication facilities required to manufacture its advanced processors.

It follows a fabless business model.

This means it focuses on chip architecture, design, software, and product development while relying on external manufacturing partners.

One of its key manufacturing partners is TSMC (Taiwan Semiconductor Manufacturing Company).

Other important suppliers provide memory, packaging, assembly, testing, and related components.

For example,

• TSMC and Samsung manufacture semiconductor wafers.
• SK Hynix, Micron, and Samsung supply memory.
• Advanced packaging processes help integrate complex semiconductor components.
• Manufacturing and assembly partners help produce complete systems.

Why is this important?

Modern AI accelerators are extremely complex products.

Their production depends on several specialised companies and manufacturing processes.

NVIDIA's success therefore depends not only on its engineering teams but also on its ability to secure manufacturing capacity, memory, advanced packaging, and other critical components.

This also creates supply-chain risks.

A shortage or delay in one part of the process can affect the availability of entire AI systems.

11. NVIDIA's Biggest Challenges

Despite its position in the AI industry, NVIDIA doesn't operate without competition or risk.

There are several challenges worth watching.

1. Competition

NVIDIA competes with companies developing alternative AI computing hardware.

Some important competitors include

• AMD, with its Instinct accelerator family.
• Intel, with its AI and data-centre products.
• Huawei, with its Ascend processors.
• Other semiconductor companies developing specialised AI hardware.

But competition doesn't come only from traditional chip manufacturers.

Some of NVIDIA's biggest customers are also developing their own chips.

For example,

• Google develops its Tensor Processing Units (TPUs).
• Amazon develops Trainium and Inferentia.
• Microsoft develops its own AI accelerator technology.

Why would customers do this?

Because relying entirely on one supplier can be expensive and risky.

Custom chips may help cloud providers optimise performance, manage costs, and reduce dependence on external hardware suppliers.

NVIDIA's challenge is to remain attractive even as customers develop alternatives.

2. The CUDA Dependency

CUDA is one of the greatest strengths of NVIDIA, but it also creates a challenge.

Developers and companies have invested years building software around its ecosystem.

That makes switching platforms difficult.

However, alternatives and compatibility layers are developing.

If competing hardware platforms become easier to program and offer strong software support, NVIDIA could face more pressure.

Its long-term advantage depends on continuously improving both hardware and software.

3. US–China Technology Restrictions

AI chips have become an important part of international technology policy.

The United States has introduced export controls that restrict the sale of certain advanced semiconductor products to China.

These restrictions affect NVIDIA's ability to sell particular products in that market.

The rules have changed over time, including conditional licensing policies for some products.

For example, the US Commerce Department announced a case-by-case licensing review policy for certain H200 exports in January 2026.

This creates uncertainty for NVIDIA.

China is a significant technology market, but geopolitical restrictions can affect which products the company is allowed to sell and to whom.

China is also investing in domestic semiconductor alternatives.

4. Manufacturing and Supply-Chain Risks

NVIDIA depends on a global network of suppliers.

Advanced chips require specialised manufacturing processes, high-bandwidth memory, and sophisticated packaging.

A shortage of advanced packaging capacity or manufacturing delays can affect how quickly NVIDIA can deliver its products.

Its reliance on Asian semiconductor manufacturing also exposes the company to geopolitical and logistical risks.

5. The Cost of AI Infrastructure

AI systems are expensive to build and operate.

They require

• High-performance processors.
• Large amounts of memory.
• Data-centre facilities.
• Electricity.
• Cooling systems.
• Networking infrastructure.

As AI adoption grows, companies must decide whether the benefits justify these costs.

If customers cannot generate enough value from their AI investments, demand for expensive infrastructure could slow down.

This is an important risk because NVIDIA's growth depends heavily on continued investment in AI computing.

12. What NVIDIA's Story Teaches Us

NVIDIA's journey offers some interesting lessons beyond semiconductor engineering.

1. Think beyond the original use case.

GPUs were developed to accelerate graphics.

But their ability to perform parallel calculations made them useful for scientific computing, machine learning, and AI.

A technology's most important application isn't always the one it was originally designed for.

2. Software can be as important as hardware.

CUDA helped turn NVIDIA's hardware into a broader computing platform.

The company didn't just sell processors. It built an ecosystem that developers could build upon.

3. Timing matters, but preparation matters too.

NVIDIA didn't suddenly become an AI company when ChatGPT became popular.

It had spent years investing in GPU computing, software, and developer tools.

When AI demand accelerated, much of the necessary foundation was already in place.

4. Building an ecosystem creates long-term advantages.

Hardware performance matters, but so do documentation, libraries, developer experience, compatibility, and community adoption.

A strong ecosystem can become a significant competitive advantage.

A technology built to make video games look better became part of the infrastructure powering the next generation of artificial intelligence.

That's quite a journey.

NVIDIA #ArtificialIntelligence #GPUs #Technology #SoftwareEngineering